This paper is organized as follows: Section II describes the NHIS data and the estimation framework. Section III presents the results, and Section IV offers a discussion and concludes.21
In the NHIS a sample adult in each household is asked: “Which of the following best represents how you think of yourself?” Response options for women include: 1: Lesbian or gay; 2: Straight, that is, not lesbian or gay; 3) Bisexual; 4) Something else; 5) I don’t know the answer; and 7) Refused.22 Approximately 2-3 percent of individuals 18 and older self-identified as gay, lesbian, or bisexual in each wave of the NHIS (Ward et al. 2013). This is similar to other large population-based surveys in the UK, US, and Canada (Joloza et al.
2010).
Individuals are also asked about their employment status, including whether they work full-time (defined as 35 hours or more per week). We also observe total earnings before taxes and deductions from all jobs and businesses in the prior calendar year which we define as annual earnings.23 In addition to the critical questions on sexual orientation and earnings, the NHIS includes standard demographic characteristics such as sex, age, race/ethnicity, educational attainment, partnership/marital status, and the presence of children in the
22 Response options for men were similar except they did not refer to ‘lesbian’. Note that individuals who responded ‘something else’ or ‘don’t know’ were further probed about the nature of those responses. These response are not included in the NHIS public use file, however, so we do not make use of them. The NHIS is a face-to-face survey with computer-aided personal interviewing (CAPI). Pilot testing by the National Center for Health Statistics showed no significant difference in sexual orientation responses by whether individuals were surveyed using CAPI or audio computer-assisted self-interviewing (ACASI). The sexual orientation question is asked in an ‘Adult Selected Items’ module that contains other questions deemed to be sensitive.
23 Approximately 16 percent of individuals who are employed full time have missing data on earnings, which is fairly standard in surveys of this type. The NHIS imputes income for these individuals, but we restrict attention to individuals who gave a non-imputed response to the earnings question.
household. We restrict attention to individuals ages 25 to 64 to focus on individuals most likely to have completed their education.24
We first estimate the relationship between sexual orientation and employment by estimating linear probability models separately by sex.25 These models take the form:
(2.1) EMPLOYEDi = α + β1Xi + β2(GAY/LESBIAN)i + β3(BISEXUAL)i + εi
where EMPLOYED is an indicator variable for having any employment or having full-time employment, depending on the model. X is a vector of demographic and job variables that (depending on the model) include: age and its square; education dummies (bachelor’s degree or more, associate degree, some college, less than high school, don’t know education, and refused education, with high school degree as the excluded category); race dummies (black only, American Indian or Alaskan Native only, Asian only, race group not releasable, and multiple race, with white as the excluded category);26 a dummy variable for Hispanic ethnicity;
relationship status dummies (widowed, divorced, separated, partnered, and marital
24 In results not reported here but available upon request, we find that lowering our minimum age in the sample to 18 returns similar results.
25 We drop a small share of observations that did not provide a valid employment status response.
26 The race of NHIS respondents may be withheld due to respondent confidentiality or other
status missing, with never married as the excluded category)27; region dummies (Midwest, South, and Northeast, with West as the excluded category); and the presence of children in the household (indicators for the presence of children ages zero to five years old and children ages six to seventeen years old). We also include survey wave dummies and month of interview dummies in all models.
Note that in this model the relevant excluded category for sexual orientation is composed of individuals who report a heterosexual orientation.28 We estimate standard errors robust to heteroscedasticity.
To assess the relationship between sexual orientation and annual earnings we estimate earnings models separately for males and females among the sample of full-time workers, following the prior literature. These models take the form:
(2.2) LOG EARNINGSi = α + β1Xi + β2(GAY/LESBIAN)i + β3(BISEXUAL)i + εi
27 Partnership is based on a dummy variable indicating the person is in any type of partnership (married or living with a partner). This accounts for the fact that legal access to same-sex marriage for sexual minorities in our sample was not universal throughout the sample period under study. Of course, individuals can still describe themselves as ‘married’ even if they are not legally married, though we have no way of identifying these individuals, regardless of the sexual orientation of the respondent.
28 In all models we separately include dummy variables for people who refused to provide a response to the sexual orientation question, or who reported ‘something else’ or ‘I don’t know’.
Demographic characteristics for these respondents are reported in Appendix B Table 2.4 and reveal that both males and females across these groups tend to be less educated, are less likely to be partnered, and are less likely to have children compared to self-identified heterosexuals. The coefficients on these indicators in the earnings regressions are reported in Appendix B Table 2.5.
where all variables are as described above. In these models we also add to the X vector: the number of years of job tenure at the current firm (and its square); a series of 26 occupation and 24 industry dummy variables; firm size categories;
and dummy variables for the sector of employment. The earnings models also include a dummy variable for whether the respondent’s personal earnings or job tenure responses were topcoded.29